Hacker News new | ask | show | jobs
by hooloovoo_zoo 2677 days ago
I think you should at least release a small portion of the training data (e.g. anything recycling related) so people can measure to what extent the model is generating new sentences and to what extent it's just regurgitating training data.